AI Does My Design Work. It Doesn't Touch the Room.

Ask a room full of facilitators whether they use AI, and you'll feel the temperature drop. Half the room hesitates, like admitting to a tool is admitting to skipping the craft. I don't feel that hesitation. I use AI constantly for design work. I keep it out of the conversation itself, on purpose.

That split is the whole argument.

The Sparring Partner

Most of the work behind a strong session happens before anyone walks into the room. That's thinking work, and a sharp sparring partner changes it. I use AI to generate variants of a structuring invitation, ten different ways to ask the same question, so I can pick the sharpest one instead of settling for the first phrasing that came to mind. I use it to pressure-test a string: does this sequence hold, where's the gap, what's missing between divergence and choice?

It also gives me a first pass at timing. How long does this arc take, does it fit inside half a day? Practice keeps correcting that estimate downward. A real group stretches, hesitates, and wanders in ways no schedule accounts for, so I build in more buffer than any model predicts. What makes AI useful here is that it never tires of "what if we structured this differently?" Asked ten times in a row, that question is exactly what good design needs.

None of this replaces the craft. It sharpens it. I used to sit alone with a blank page. Now I have a partner that forces me to defend every choice before I make it. The decision stays mine. The options I'm choosing from are better.

Where the Model Goes Quiet

Then you hit the edge, and for me it's sharp. The active ingredient in Liberating Structures isn't on the page. It's what happens between people in the room: the silence after an invitation, the moment someone says out loud what they'd never said before, the way a group learns from itself as it goes. You don't generate that. You provoke it and hold it.

AI doesn't know your group. It doesn't feel the room. It doesn't know that the woman in the back has already started to speak twice and swallowed both attempts. It can't hear hesitation in an answer, and it can't tell a productive silence from an uncomfortable one. That's the whole job of facilitating. AI can help me write a sharper invitation. It cannot carry the silence that follows it.

What About the Participants?

The next question is always close behind: should participants use AI during a session? I'm holding back on that one. The heart of most structures is that people think alone, in silence, before they share anything, and the first step of 1-2-4-All is the clearest example. Outsource that thinking to a model and you've removed the active ingredient. What a group surfaces has to come from the group. That's the whole point: the answer is already in the room.

I want to be honest here too: I'm not certain. I haven't tested this systematically enough to be definitive, and I can imagine contexts where AI as a thinking tool inside a session adds real value, prepping materials, or quickly organizing what a large group has just produced in a breakout. That needs more experiment before I'll commit to a verdict. For now, I keep the conversation itself as human work. Sorting what a model can take over from what it can't takes the same kind of diagnosis as telling a complicated problem apart from a complex one.

AI is a tool for the work around the conversation. The conversation belongs to the people in the room. Keep that line sharp, and the technology strengthens the practice instead of thinning it out. The day I start valuing the invitation more than what happens after it, I've lost the line.

Do you use AI in your prep? And where do you draw the line between what you hand off and what stays human work?